STYLE: A technology-independent approach to statistical design

Despite their indisputable benefits, parametric yield prediction/optimization techniques are often not routinely used by industry because of prohibitive costs and/or lack of generality. To overcome these obstacles, we present a new statistical technology independent parametric yield approach which is computationally inexpensive and generally applicable, unlike previous inexpensive methods which limit the user to specific technologies, device models, or circuits. This general applicability and cost efficiency is maintained through four factors. First, this approach interfaces with a compiler-methodology based simulator which establishes complete model independence. Second, critical parameters are not predetermined; an efficient screening procedure minimizes the dimensionality of the input parameter space so as to formulate a parsimonious nonparametric regression model. Third, because of the flexibility of this nonparametric model, the capability exists to map any input parameters to an output performance, not just those with a linear relationship. Finally, a waveform analysis capability completely automates the approach, thereby enabling turn-key operation. An ECL bandgap regulator circuit illustrates the benefits of this approach.